SD WebUI Tag Autocomplete: Booru Tag Completion Inside AUTOMATIC1111

What problem tag autocomplete solves, and for whom

Stable Diffusion checkpoints trained on booru-style datasets respond to tags such as character names, artist names and composition keywords. The catch is spelling. A prompt that writes a tag slightly differently from the dataset annotation moves the model toward a different concept, and the booru tag space is large and full of near-duplicates. Tag Autocomplete exists to remove that guesswork: it is an extension for AUTOMATIC1111's Stable Diffusion web UI that shows completion hints for recognized tags from image booru boards such as Danbooru while you type in the prompt field.

The audience is narrow and specific. It is for people who already prompt with booru tags, or who want to, and who run the AUTOMATIC1111 web UI. The README's own reasoning is that most custom Stable Diffusion models were trained on this information or merged with ones that were, so exact tags can improve composition and consistency even when the model has a photorealistic style. If you prompt in full sentences or use a different front end, the extension has little to offer you.

One warning comes with the default data: the README states that the default tag lists contain NSFW terms and asks users to use them responsibly. This is not an opt-in pack you fetch separately; it is what ships.

How completion is triggered: tags, __, < and chants

The extension watches the prompt textarea and matches what you have typed so far against several separate sources. Ordinary typing matches against the bundled tag lists: Danbooru and e621, described in the README as the top 100k tags by post count as of November 2022. Those lists live in the repository's tags/ directory, so the vocabulary is fixed at whatever the released version carries rather than fetched live from the boorus.

Other sources use trigger characters. Typing two underscores, __, starts wildcard completion: it first lists your wildcard files, and after you pick one it lists the replacement options inside that file. Wildcards are searched for in every extension folder as well as the legacy scripts/wildcards folder, and nested folders are supported, so wildcards from several extensions can be combined in one list. Typing < starts extra network completion for embeddings, hypernetworks, LoRA and LyCORIS or LoHA. By default all types are mixed together; the README lists these prefixes for narrowing the list: <e: for embeddings, <l: for LoRA and LyCORIS with <lora: and <lyco: as the long forms, and <h: or <hypernet: for hypernetworks.

The extension also supports chants, described as a custom format for longer prompt presets, plus a single customizable extra tag file. For extra networks that have preview images, the popup shows a small live preview window beside the regular list, which stays hidden for normal tags or when no preview image was found. That preview is the feature that saves the most time in practice, because LoRA filenames are frequently opaque.

Installing a1111-sd-webui-tagcomplete and a first real prompt

The README gives two installation paths. The first is the web UI's own extension browser: open the Extensions tab, switch to the Available sub-tab, click Load from, find Booru tag autocompletion in the list, and click Install. The README notes the extension was one of the first available, so sorting oldest first puts it near the top, and that you can use Ctrl+F to search the page text instead. The second path is a manual clone into the extensions folder, where the second argument names the destination folder and can be anything you like:

git clone "https://github.com/DominikDoom/a1111-sd-webui-tagcomplete.git" extensions/tag-autocomplete

A pre-packaged version is also offered on the Releases page. After installing, restart the web UI. A first useful test is to type a few letters of a common composition tag in the prompt box and watch for the popup; keyboard navigation is supported, so you can move through the list and accept an entry without leaving the textarea.

The second test exercises the extra network path, since that is where the trigger characters matter. Type the less-than sign followed by the LoRA prefix and the start of a filename, for example <l: followed by a few letters of one of your LoRA files. You should see matching LoRA and LyCORIS files from your models folder, and if any of them have preview images, a small preview window next to the list. If the list stays empty, the extension is not finding your model folder, which is the most common first-run problem rather than a bug in the completion itself.

Where it breaks: stale tag lists, missing folders and NSFW defaults

The bundled tag data is a snapshot, not a feed. Danbooru and e621 are described as the top 100k tags by post count as of November 2022, so tags coined after that date simply are not in the list, and the release notes for 3.2.0 mention tag updates, which tells you refreshing them is a periodic release activity rather than something the extension does for you. If you work with a niche character or a recently popular artist, expect to type that tag manually and consider the extra file for your own recurring terms.

The wildcard and extra network sources depend on where files live on disk. Wildcards are looked up across every extension folder and scripts/wildcards, which is generous but also means a stray file in an unrelated extension can appear in your completion list. Extra networks depend on the web UI's own model folders being where the extension expects them; the README does not document an override for those paths, so a non-standard layout is a plausible source of an empty list.

Finally, the default NSFW content is a real deployment consideration. On a shared machine or a streamed session, the completion popup will surface explicit terms as you type. The README does not describe a switch that strips them from the shipped lists, so the practical mitigation is settings-level filtering or not installing it on that machine.

Compared with all-in-one prompt tools and ComfyUI nodes

The closest alternative in the AUTOMATIC1111 ecosystem is a prompt all-in-one style extension, which bundles autocomplete together with prompt editing features such as attention weighting, token counting and syntax highlighting in one panel. The difference in approach is scope: Tag Autocomplete does one job, matching your text against tag, wildcard and model filename lists, and leaves prompt syntax manipulation to the web UI itself. An all-in-one tool gives you one place to configure everything, at the cost of a larger surface that you partly do not need. If you already run several prompt extensions, adding a single-purpose one is easier to reason about when something misbehaves.

On the ComfyUI side there is no equivalent here. The README describes an extension for AUTOMATIC1111's web UI, and the extra network completion is tied to that UI's model folder conventions and its Extra Networks menu, whose card preview images the extension reuses. A ComfyUI node graph has a different prompting surface, so this project is not the tool for it. The same applies to other web UI forks: the extension depends on the host UI's extension API and folder layout, and the README does not claim support beyond the AUTOMATIC1111 web UI.

Maintenance, licence and the cost of upgrading

The repository is not archived, and its last push was on 2026-07-01, so there is ongoing activity. Releases are infrequent rather than continuous: 3.1.0 in August 2024, 3.2.0 in January 2025 and 3.3.0 in May 2025. That cadence matters for planning. Tag data refreshes and compatibility fixes arrive with releases, so a pinned older version will drift from the current booru vocabulary.

The project is MIT licensed. In practice that means you can read, modify and redistribute the code, including inside a commercial workflow, provided the licence text and copyright notice travel with it; the bundled tag lists come from Danbooru and e621, and the README does not discuss the terms of those upstream datasets. That is worth checking yourself if you redistribute the tags/ directory, since the repository's MIT licence covers the project's own code and the README is silent on the data's provenance terms.

Upgrade cost is low in the normal case. Install through the extension list and updates appear in the Extensions tab; a manual clone is updated with git pull in the tag-autocomplete folder. The risk is not migration but drift: a version bump can change tag lists and completion behaviour, so if you have tuned your workflow around specific entries, re-check the popup after updating. The README does not document a rollback procedure, so keep the release you are on until you have confirmed the new one behaves.

Editorial conclusion

Adopt it if you write prompts in booru tag style and want the exact Danbooru or e621 spelling, wildcard names and LoRA filenames offered as you type. Skip it if your prompts are plain sentences, since the completion list keys off tag vocabulary rather than natural language, and skip it on any UI other than AUTOMATIC1111's, because the README describes an extension for that web UI only. Before relying on it, check the tags/ folder in your clone to confirm which tag lists shipped, and test one wildcard file and one LoRA with <l: to see whether your folder layout is picked up.

Frequently asked questions

Official sources

1. DominikDoom/a1111-sd-webui-tagcomplete on GitHub
2. Issues
3. License: MIT
4. README
5. Releases